Refactoring the Agent: From Hardcoded Logic to Unified LangGraph & MCPs - Maor Oz (Hebrew) скачать в хорошем качестве

Refactoring the Agent: From Hardcoded Logic to Unified LangGraph & MCPs - Maor Oz (Hebrew)

How do you evolve from a fragile, hardcoded script into a scalable agent architecture? At DoorLoop, we began with almost no AI context, stitching together limited workflows without tools. It quickly became clear that this approach could not scale. In this session, we will explore the real engineering journey of building an AI platform, tracing the evolution from early RAG experiments to fully stateful, production-grade agents. We’ll cover the steep learning curve of adopting LangChain, the early limitations of tool-less and prompt-only workflows, and the pivotal shift to the ReAct paradigm. Finally, we’ll look ahead to the future of agent scalability, focusing on how the Model Context Protocol (MCP) enables clean, standardized connections between agents and internal tools, unlocking a more modular, extensible, and maintainable AI architecture. Maor Oz is an AI Technical Lead at DoorLoop. Over the past two years, Maor was a Technical Lead focused on strengthening the company’s infrastructure and developer experience. Six months ago, he transitioned into an AI Technical Lead role, where he now leads DoorLoop’s initiative to introduce and scale multiple AI agents across the company. In his free time, you can find Maor playing various instruments, creating music and playing video games. For more meetups, join our Meetup group - https://www.meetup.com/honeybook-engi... And check our Tech Blog - https://honeybook.engineering/